aprender-serve 0.65.2

Pure Rust ML inference engine built from scratch - model serving for GGUF and safetensors
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301


#[test]
fn test_generate_with_cache_repetition_penalty() {
    let config = create_test_config();
    let transformer = AprTransformer::new(config);

    let gen_config = GenerateConfig {
        max_tokens: 5,
        temperature: 1.0,
        top_k: 0,
        top_p: 1.0,
        repetition_penalty: 1.5, // Strong repetition penalty
        trace: false,
        stop_tokens: vec![],
        cancel: crate::generate::CancelToken::never(),
    };

    let result = transformer.generate_with_cache(&[0, 1, 2], &gen_config);
    assert!(result.is_ok());
}

// ============================================================================
// Part 9: AprKVCache Edge Cases
// ============================================================================

#[test]
fn test_kv_cache_with_gqa_config() {
    let config = AprTransformerConfig {
        architecture: "test".to_string(),
        hidden_dim: 128,
        num_layers: 2,
        num_heads: 8,
        num_kv_heads: 2, // GQA
        vocab_size: 50,
        intermediate_dim: 256,
        context_length: 64,
        rope_theta: 10000.0,
        eps: 1e-5,
            eos_token_id: None,
    ..Default::default()
    };

    let cache = AprKVCache::new(&config);

    assert_eq!(cache.num_kv_heads(), 2);
    assert_eq!(cache.head_dim(), 128 / 8);
    assert_eq!(cache.capacity(), 64);
}

#[test]
fn test_kv_cache_append_all_layers() {
    let config = create_test_config();
    let mut cache = AprKVCache::new(&config);

    let kv_size = config.num_kv_heads * (config.hidden_dim / config.num_heads);
    let k = vec![1.0f32; kv_size];
    let v = vec![2.0f32; kv_size];

    // Append to all layers (last layer auto-advances)
    for layer in 0..config.num_layers {
        cache.append(layer, &k, &v);
    }
    // No advance() needed - append() auto-advances on last layer

    // Cache length is counted per position, not per layer
    // After appending once to each layer at position 0
    assert_eq!(cache.len(), 1);
}

#[test]
fn test_kv_cache_multiple_positions_all_layers() {
    let config = create_test_config();
    let mut cache = AprKVCache::new(&config);

    let kv_size = config.num_kv_heads * (config.hidden_dim / config.num_heads);

    // Simulate 5 positions
    for pos in 0..5 {
        let k = vec![(pos + 1) as f32; kv_size];
        let v = vec![(pos + 10) as f32; kv_size];

        for layer in 0..config.num_layers {
            cache.append(layer, &k, &v);
        }
        // No advance() needed - append() auto-advances on last layer
    }

    assert_eq!(cache.len(), 5);

    // Verify each layer's cache
    for layer in 0..config.num_layers {
        let (k_cache, v_cache) = cache.get(layer);
        assert_eq!(k_cache.len(), 5 * kv_size);
        assert_eq!(v_cache.len(), 5 * kv_size);
    }
}

// ============================================================================
// Part 10: AprTransformerLayer Edge Cases
// ============================================================================

#[test]
fn test_layer_with_all_optional_biases() {
    let hidden_dim = 64;
    let intermediate_dim = 128;

    let mut layer = AprTransformerLayer::empty(hidden_dim, intermediate_dim);

    // Add all optional biases
    layer.attn_norm_bias = Some(vec![0.01; hidden_dim]);
    layer.qkv_bias = Some(vec![0.01; hidden_dim * 3]);
    layer.attn_output_bias = Some(vec![0.01; hidden_dim]);
    layer.ffn_gate_weight = Some(vec![0.1; hidden_dim * intermediate_dim]);
    layer.ffn_gate_bias = Some(vec![0.01; intermediate_dim]);
    layer.ffn_up_bias = Some(vec![0.01; intermediate_dim]);
    layer.ffn_down_bias = Some(vec![0.01; hidden_dim]);
    layer.ffn_norm_weight = Some(vec![1.0; hidden_dim]);
    layer.ffn_norm_bias = Some(vec![0.01; hidden_dim]);

    // Count parameters should include all optionals
    let params = layer.num_parameters();
    assert!(params > hidden_dim * 3 * hidden_dim); // More than just qkv
}

#[test]
fn test_layer_empty_gqa_various_ratios() {
    // Test various GQA ratios
    for (num_heads, num_kv_heads) in [(8, 4), (8, 2), (16, 4), (32, 8)] {
        let hidden_dim = num_heads * 16; // head_dim = 16
        let intermediate_dim = hidden_dim * 4;

        let layer =
            AprTransformerLayer::empty_gqa(hidden_dim, num_heads, num_kv_heads, intermediate_dim);

        let head_dim = hidden_dim / num_heads;
        let kv_dim = num_kv_heads * head_dim;
        let qkv_out_dim = hidden_dim + 2 * kv_dim;

        assert_eq!(layer.qkv_weight.len(), hidden_dim * qkv_out_dim);
    }
}

// ============================================================================
// Part 11: Forward Pass with Various Configurations
// ============================================================================

#[test]
fn test_forward_single_layer_single_head() {
    let config = AprTransformerConfig {
        architecture: "test".to_string(),
        hidden_dim: 32,
        num_layers: 1,
        num_heads: 1,
        num_kv_heads: 1,
        vocab_size: 50,
        intermediate_dim: 64,
        context_length: 64,
        rope_theta: 10000.0,
        eps: 1e-5,
            eos_token_id: None,
    ..Default::default()
    };

    let transformer = AprTransformer::new(config);
    let result = transformer.forward(&[0, 1, 2, 3, 4, 5, 6, 7]);
    assert!(result.is_ok());
}

#[test]
fn test_forward_many_layers() {
    let config = AprTransformerConfig {
        architecture: "test".to_string(),
        hidden_dim: 32,
        num_layers: 12, // Many layers
        num_heads: 2,
        num_kv_heads: 2,
        vocab_size: 50,
        intermediate_dim: 64,
        context_length: 64,
        rope_theta: 10000.0,
        eps: 1e-5,
            eos_token_id: None,
    ..Default::default()
    };

    let transformer = AprTransformer::new(config);
    let result = transformer.forward(&[0, 1]);
    assert!(result.is_ok());
}

#[test]
fn test_forward_with_cache_many_layers() {
    let config = AprTransformerConfig {
        architecture: "test".to_string(),
        hidden_dim: 32,
        num_layers: 8,
        num_heads: 2,
        num_kv_heads: 2,
        vocab_size: 50,
        intermediate_dim: 64,
        context_length: 64,
        rope_theta: 10000.0,
        eps: 1e-5,
            eos_token_id: None,
    ..Default::default()
    };

    let transformer = AprTransformer::new(config.clone());
    let mut cache = AprKVCache::new(&config);

    for pos in 0..10 {
        let result = transformer.forward_with_cache(pos as u32, &mut cache, pos);
        assert!(result.is_ok());
    }
}

// ============================================================================
// Part 12: Predict Next Edge Cases
// ============================================================================

#[test]
fn test_predict_next_returns_valid_token() {
    let config = create_test_config();
    let vocab_size = config.vocab_size;
    let transformer = AprTransformer::new(config);

    for tokens in [vec![0], vec![0, 1], vec![0, 1, 2, 3, 4]] {
        let result = transformer.predict_next(&tokens);
        assert!(result.is_ok());
        assert!((result.expect("test value should be present") as usize) < vocab_size);
    }
}

#[test]
fn test_predict_next_deterministic() {
    let config = create_test_config();
    let transformer = AprTransformer::new(config);

    // Same input should produce same output (no randomness in predict_next)
    let r1 = transformer.predict_next(&[0, 1, 2]).expect("test value should be present");
    let r2 = transformer.predict_next(&[0, 1, 2]).expect("test value should be present");
    assert_eq!(r1, r2);
}

// ============================================================================
// Part 13: Num Parameters and Memory Size
// ============================================================================

#[test]
fn test_num_parameters_with_optionals() {
    let config = create_test_config();
    let mut transformer = AprTransformer::new(config.clone());

    let base_params = transformer.num_parameters();

    // Add optional biases
    transformer.output_norm_bias = Some(vec![0.0; config.hidden_dim]);
    transformer.lm_head_bias = Some(vec![0.0; config.vocab_size]);

    let with_optionals = transformer.num_parameters();

    // Should be larger with optionals
    assert!(with_optionals > base_params);
    assert_eq!(
        with_optionals - base_params,
        config.hidden_dim + config.vocab_size
    );
}

#[test]
fn test_memory_size_consistency() {
    let config = create_test_config();
    let transformer = AprTransformer::new(config);

    let params = transformer.num_parameters();
    let memory = transformer.memory_size();

    // Memory = params * 4 (f32 = 4 bytes)
    assert_eq!(memory, params * 4);
}

// ============================================================================
// Helper Functions
// ============================================================================

fn create_test_config() -> AprTransformerConfig {
    AprTransformerConfig {
        architecture: "test".to_string(),
        hidden_dim: 64,
        num_layers: 2,
        num_heads: 4,
        num_kv_heads: 4,
        vocab_size: 100,
        intermediate_dim: 256,
        context_length: 128,
        rope_theta: 10000.0,
        eps: 1e-5,
            eos_token_id: None,
    ..Default::default()
    }
}